Grayscale medical imaging, such as X-ray images, has been standardized for many years. The Grayscale Standard Display Function (GSDF) defined by NEMA in 1998 describes exactly the transfer curve for a medical grayscale display function. The result of applying this function on the image and the display is a perceptually linear representation of the grayscale image. This brings for this field of medical imaging the great benefit that the visibility of medical features is uniform over the complete range of digital driving levels (DDL). When the grayscale image is not represented in a perceptually linear way, errors may occur. For example, a chest nodule could appear or disappear completely because its grayscale value is just some DDL's too high or too low, resulting in a difference with its surroundings that is below the threshold of visibility. This kind of error is not acceptable for medical imaging. The perceived threshold or Just Noticeable Difference (JND) is derived from Barten's work on the Contrast Sensitivity Function (Barten, 1992).
In recent years medical imaging is evolving more and more from pure grayscale images to color images. Until now, color medical imaging has not been standardized, although the situation with colored images case is a bit more complex. Depending on the specific field of medicine, there may be other requirements for the representation of colors. For surgery and examination using for instance endoscopes, an exact representation of colors is a prerequisite. The endoscope combined with the display can be considered as an extension of the doctor's eyes and hence should present an image that is the same as would be provided to the doctor. The same can be held for the interpretation of wound photographs used in tele-medicine, where the color is giving an indication if a wound is healing.
The situation is different for the emerging markets of digital pathology or quantitative imaging. For this kind of images it is of great importance, similar to the situation depicted for grayscale images, that the doctor is able to discover relevant medical features in the images. To facilitate the discovery it is important to visualize especially the differences between the features and the background of the image. Hence distinguishability can be more important than a perfectly truthful image.
In a conventional digital image processing chain for pathology, the display is conventionally not considered as an essential part to optimize the detectability of the features in the scanned slides. The approach so far is to represent the colors in exactly the same way as how the pathologist would perceive them when looking through the microscope. To obtain this, the scanned slide is for instance saved in the sRGB color space and the display is assumed to be sRGB calibrated. In the best case ICC profiles can be used to take into account the gamut of the actual display or a specific calibration method is applied to guarantee accurate color reproduction, see for example “WO2013025688 SYSTEM AND APPARATUS FOR THE CALIBRATION AND MANAGEMENT OF COLOR IN MICROSCOPE SLIDES”.
This approach has some flaws. First of all, what is the “correct” color? The colors that are perceived when using a microscope depend on the spectrum of the light source of the microscope. Thus, a slide will look different from microscope to microscope or from set up to set up. In addition, hospitals or laboratories often have their own procedures for preparing slides and to perform the staining Although more or less the same procedure is used in different labs, the intensity of the staining can vary significantly. To make the situation even more complex, after scanning the slides the colors can differ even more depending on the scanner used. Different scanners with the same illumination can produce images with different colors. Therefore it is not advisable to rely on the exact representation of colors for digital pathology applications.
In quantitative medical imaging, the results of calculations are visualized using pseudo colors on top of other medical images or as images on their own. Because these colors are calculated, it is possible to define a color space in which the image is rendered, for instance sRGB, and by using a display and the correct ICC profiles, the calculated colors can be quite accurately visualized.
However, in such images often there is only a small amount of the scale that is represented by one primary color such as red, whereas another primary color such as green can represent the biggest range of the quantitative values, making it difficult to distinguish the colors in this scale. Using a perceptually linear color scale can help optimize the visualization of the quantitative colors and reveal potentially hidden details in the image. This can only be realized when taking into account the gamut of the display used for the visualization of the image. In both digital pathology and quantitative imaging it is critical to optimally visualize the differences between the features and the background. Therefore, with a similar reasoning one can conclude that digital pathology images may be better interpreted on a perceptually linear color display.
Calibrating a display in such a way that it is perceived as being linear may involve using a perceptually uniform color space. One such color space is proposed in “Toward a Unified Color Space for Perception-Based Image Processing”, Ingmar Lissner and Philipp Urban, IEEE Transactions on Image Processing, (Volume: 21, Issue: 3), 4 Aug. 2011 ISSN: 1057-7149. Their “perceptually uniform” and “hue linear” color space is called LAB2000HL (including variations optimized for color difference metrics other than ΔE2000) and is derived from CIELAB and ΔE2000. In this paper reference to “perceptually uniform” means that ΔE2000 within LAB2000HL is a only Euclidean distance locally and it is shown that it is impossible to design a color space in which ΔE2000 is a true Euclidean distance other than locally. The paper discloses iterative adjustment of the color grid points on equi-luminance planes, while enforcing some other constraints including hue-linearity, which causes some loss in perceptual uniformity.
Another perceptually linear color space contender, UP Lab (http://www.brucelindbloom.com/UPLab.html) does a better job for sRGB blue primary but has problems for green and red. Without being limited by theory, these problems may be due to the fact that both UP Lab and LAB2000HL separate luminance and chrominance at the outset while there is evidence in the literature that the two may not be treated separately in constructing a perceptually uniform color space.
For a color display calibration suited for medical applications, there is a need to find a method of distributing color points across a full display gamut in a perceptually uniform manner while preserving full contrast and color saturation in the calibrated display and without the problems mentioned above with the prior art methods.